Multi-attribute Vehicle Recognition with Feature Fusion Using Muti-Gate Unit

Vehicle multi-attribute recognition is a critical technology in modern transportation systems. It can help automatically recognize and classifies vehicles based on attributes such as manufacturer, model, color, and type. This information is helpful for a variety of downstream applications, including traffic surveillance, intelligent transportation, and vehicle tracking. Currently, the benchmark performance of image recognition models continues to improve with the ongoing development of deep learning technology. However, most recognition models still have room for improvement in tasks involving the simultaneous recognition of multiple attributes. In this paper, we propose a new method for enhancing the performance deep learning models in multi-attribute recognition tasks. Our method is inspired by the idea of Muti-Gate Mixture of Experts (MMoE) network, which can effectively mix models of different scales of features for the recognition of different attributes. We also design a loss function that combines information entropy and contrastive loss to enhance the accuracy of fine-grained image recognition in multi-task scenarios with limited datasets. We apply our method to a vehicle multi-attribute recognition model and demonstrate through results the effectiveness of our proposed method in improving deep learning models for multi-attribute recognition tasks.